Tasks

​

What Tasks are

Tasks turn your agent from a passive chatbot into something that keeps working when you close the tab. Two kinds share one board:

• Schedules run on their own — a weekly site review, an hourly inbox check. They repeat.
• Cards are single pieces of work. Your agent files them when it finds something worth doing; you decide what actually happens.

​

Availability

Tasks are available from Free and there is no task-count ceiling. Message credits and rate limits govern the work tasks run, rather than the number of cards or schedules you can keep. Accounts with no plan cannot create or run Tasks.

​

The board

Five columns, and work moves left to right: Drag a card to move it. Above the board, a Schedules strip lists the recurring runs — they have no column because they never finish.

​

Autonomy

How much your agent may do without you, set per Site and overridable per task: Publishing is always yours. No autonomy rung lets an agent put changes live.

​

Site reviews

Every Site gets one review schedule. It reads your pages, screenshots them, checks what visitors have been asking, and files what is worth improving — with its reasons. It changes nothing on its own. The visitor half is the part only your agent can do. Anyone can tell you a page needs a clearer call to action; only your agent knows that six people asked what something costs and there is no pricing page. Choose the cadence in Site builder → Settings → Reviews & autonomy. Unavailable cadences stay visible with the plan that unlocks them. Hourly is a rolling 60-minute interval. Calendar cadences run at their configured calendar time. After a downgrade, a cadence the new plan does not include is changed to the fastest cadence that plan permits.

​

Creating your own

Ask in the builder conversation — “check my inbox every morning at 9” — or use the schedule control in the composer. Recurring tasks appear in the Schedules strip; one-off ones land on the board.

​

Configuration

​

Plan limits

Free and every paid plan include unlimited Tasks. Each accepted attachment used by a Task or chat costs two ordinary message credits, and task executions consume the same credits and rate limits as other Agent work.

​

Runs and history

Below the board:

• Run history — every run with status, duration and tokens used
• Activity chart — execution volume over time
• Stats — success rate, total executions, active schedules

A run that was skipped records why — outside active hours, disabled, or sitting in a column that does not run. That is the answer to “why didn’t my task run last Tuesday?”, and it is the reason run history exists separately from the conversation. Successful runs happen in your builder conversation, so the full exchange is there to read.

​

Safety

• Circuit breaker — five consecutive failures (configurable) pauses a task rather than letting it fail forever
• Plan and credit gate — Tasks stop when the plan is inactive or the account cannot fund a run
• State guard — a card only runs while it is In Progress, so a Backlog card cannot fire
• Publishing — never automatic, at any autonomy rung

​

Best practices

• Let Site reviews fill your Backlog before creating tasks by hand
• Read the agent’s reason on a card before promoting it; that is what it is for
• Start at Suggest, move a Site to Autopilot once you trust what it files
• Check Review first — it is the only column that needs you

​

Next steps